Models 1-5 include all samples of loans (N), both fully paid and defaulted. Models 6-9 for Sample 1 and Sample 2 indicate the period before and after the interest rate change both for fully paid and defaulted loans. Each number represents computed marginal coefficients and ‘*’ corresponding significance level. Each model differs depending on the explanatory variables included. McFadden R-squared and Hosmer-Lemeshow Test indicates the model’s goodness-of-fit, for the later being higher than 0.05.
N
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Borrower Assesment Grade 0,078*** 0,091*** 0,077*** 0,0833*** 0,099*** Interest rate 1,653*** 1,969*** 1,078*** 1,296*** Loan term 0,197*** 0,202*** 0,208*** 0,198*** 0,219*** 0,201*** 0,228*** 0,247*** 0,296*** Loan characteristics Purpose: Investment 0,058 0,036 0,333* -0,181 Refinancing 0,109 0,102 0,352 -0,001 Education 0,153 0,140 0,382 -0,029 Working capital 0,219** 0,191* 0,489** 0,041 Consumption 0,057 0,036 0,489** -0,234 Other 0,082 0,060 0,339* -0,078 Loan amount -0,000*** 0,000*** 0,000** 0,000 Number of characters -0,001 0,000 0,000 0,000 Base currency -0,043 -0,027 0,018 -0,091 Borrower characteristics Annual income 0,000 0,000 0,000* 0,000 Employment type: Salaried 0,108 0,138 -0,029 0,526 Self employed 0,097 0,132 -0,009 0,475 Studying 0,129 0,129 -0,045 1,123 Unemployed 0,015 0,01 -0,166 0,281 Industry: Financial services -0,087 -0,096 0,133 -0,281*
Information and com -0,091 -0,106 0,202 -0,463***
Proffesional and scientific -0,085 -0,112 0,169 -0,324**
Manufacturing -0,087 -0,09 0,274 -0,516***
Education -0,177 -0,215* 0,053 -0,351
Other services -0,027 -0,061 0,219 -0,238
Wholesale and retail -0,004 -0,044 0,170 -0,209
Public and defense -0,065 -0,076 0,125 -0,220
Human health -0,015 -0,027 0,305 -0,246
Transportation 0,254 0,266** 0,739*** -0,267
Arts and entertainment -0,103 -0,129 0,317 -0,928***
Admin and support 0,076 -0,055 0,408** -0,217
Construction 0,184 -0,168 0,415** 0,076
Agriculture -0,123 -0,153 0,162 -0,620**
Electricity 0,119 -0,075 0,329 -0,099
Accommodation and food 0,1106 0,069 0,491** -0,373
Country:
Developed -0,143 -0,108 0,003 -0,047
Developing -0,133 -0,128 0,024 -0,061
Total identifications -0,011* -0,016** -0,024** 0,001
Borrower indebtedness
Loan amount to ann. income -0,024 -0,009 -0,041 0,009
McFadden R-squared 0,096 0,113 0,119 0,097 0,138 0,086 0,152 0,112 0,190
Hosmer-Lemeshow Test 0,000 0,239 0,479 0,000 0,645 0,001 0,071 0,930 0,427
*** significant at the 1% level ** significant at 5% the level * significant at the 10% level
Full model Sample 1 Sample 2
6.
C
ONCLUSION
Bitcoin and P2P combination - bitcoin lending - is seen as an alternative financing option to traditional financial institutions due to its lower transaction costs, accessibility for non “bankable” borrowers, time-effective and transparent processes. Since borrowers and lenders are simply put together into one platform, automation reduces processing costs, which are the most important expenses in the banking industry, therefore, it provides a technological advantage for bitcoin lending. It is also attractive as a foreign currency investment due to its international lending and borrowing option, reduced risk for foreign exchange exposure and usefulness as a speculative tool, giving advantage against usual p2p lending. In addition, global diversification, accessibility of higher Return of Investment rates and lower fees distinguish bitcoin lending from p2p markets. The credit rationing problem is reduced through online lending, which explains the growth of this market. On the other hand, the information asymmetry reduction is a crucial issue in this market, since the risk is faced directly by the individual lenders while in the banking industry credit risk management is under the financial institution itself with analysts providing their expertise. Therefore, bitcoin lending platforms face steep challenges to provide quality information about its borrowers and loans in order to reduce credit risk. This can be done by obtaining the information from the platform itself as a grade assigned for each borrower or relying on third party credit scoring.
This paper analyses whether the additional information such as borrowers’ and loans’ characteristics, without the interest rate or grade in the bitcoin lending platform Bitbond can fully explain probability of default and reduce information asymmetry. An empirical study has been conducted to test the hypotheses on variables influencing probability of default. Descriptive loans’ analysis indicates that 80% of the Bitbond’s platform is concentrated within C-F credit rating borrowers, with default rates higher than 41.3% and interest rates of 28%, on average. Concentration of risky borrowers might be explained as a still developing market’s problem, attracting speculative investors and opportunistic borrowers. Since requirements to get funded are minimal, 63% of borrowers have been funded. Comparing results with the Lending Club’s research revealed that Bitbond is subjected to much higher risk, as well as returns. Logistic regressions were set up for 9 different models in order to predict the defaults. The results indicate that there is no clear relation between the grade assigned by the Bitbond and the default, i.e., 20-25% of A-B graded loans are defaulted, while the percentage rapidly increased to 43-50% within D-F grades. Thus, it indicates that lenders, especially within speculative class, are faced with uncertainty and should not fully rely on the grade as a
52
determinant of default probability. The loan term is another factor potentially explaining default – there is an extreme pattern of loans longer than 6 weeks to be subjected to more than 50% default rate. Loan amount, loan purpose as working capital, loan industry as education and transportation and total identification were found to be significant as well. However, there was no significance difference found with annual income, length of description, employment type, between country of origin or base currency choice. Additional research was made on the two subsamples due to interest rate increase and a significant difference was found between the explanatory variables. It might be caused by different borrowers’ profile after interest rate change due to their willingness to reduce information asymmetry in order to access lower nominal interest rates. In both periods nominal interest rate indicates a clear relation with default probability, however, there is no significant difference in comparison to the full grade model of goodness-of-fit. Different significant explanatory variables between interest rate or grade based models exist.
To sum up, information provided by the Bitbond platform gives the right to see loans’/borrowers’ characteristics, borrowing history, credit grades assigned and success of funding. Therefore, information asymmetry is partly reduced through the qualitative data provided. Nevertheless, bitcoin loans are exposed to higher risk, instability and default rates compared with the usual p2p lending, so lenders should be careful with their investment choice. As the research is limited, possible recommendations for the future study are to include a larger sample of data, use larger time frame and add more additional variables, like soft data, which would increase explanatory power of the models. Moreover, deep and comparative analysis with p2p lending market would provide more insights for actual risk-returns.